Falling into local optima, market prediction should not stop here

Title: Two Kites Dancing In A Hurricane

Author: 0xsmac

Source:

Repost: Mars Finance

Editor’s note: This article critically examines the current prosperity of the prediction markets. The author sharply points out that today’s prediction markets are陷入黑莓和雅虎式的「局部最优解」陷阱。Mainstream binary options-based prediction markets, while gaining massive traffic in the short term, are plagued by structural issues such as liquidity shortages and low capital efficiency. The article proposes an evolution towards a「perpetual contract」model, offering constructive deep insights for realizing a true「market of all things」.

Why do companies find themselves chasing the wrong goals? Can we fix prediction markets before it’s too late?

“Success is like a strong drink, intoxicating. Mastering the fame and praise that come with it is no easy feat. It can erode your mind, making you believe everyone around you fears and respects you, everyone desires you, and everyone’s thoughts are constantly revolving around you.” — Ajith Kumar

“The cheers of the crowd have always been the most beautiful music.” — Vin Scully

Early success is intoxicating. Especially when everyone tells you you won’t succeed, that feeling becomes even stronger. To hell with the haters—you are right, they are wrong!

But early success harbors a unique danger: you may be winning the wrong rewards. We often joke that “playing stupid games wins stupid prizes,” but in reality, the games we participate in are often evolving in real time. Therefore, the factors that helped you win in the first phase may become stumbling blocks when the game matures and you seek bigger rewards.

One manifestation of this outcome is: companies unknowingly fall into a「local optimum」. The feeling of winning is so good that it not only causes you to lose direction but also blinds you from self-awareness, making it impossible to see your true situation.

In many cases, this may just be an illusion supported by external factors (such as economic prosperity leading to an overflow of disposable income among consumers). Or, your product or service may indeed operate well within a limited scope or under certain conditions but cannot be scaled to a broader market.

The core conflict here is: to pursue the true ultimate prize (the global optimum), you need to come down from the current peak. This requires immense humility. It means making tough decisions: abandoning a core feature, completely rebuilding the tech stack, or personally overturning a mode you once thought effective. Making all this more challenging is…

Most of the time, you have to make these decisions when people (mainly investors and media) are telling you “how great you are”! Many who previously said you were wrong now rush to verify your success. This is an extremely dangerous situation because it breeds complacency precisely when you need to make radical changes.

This is exactly where prediction markets are today. In their current form, they can never achieve mass adoption. I don’t want to waste words debating whether they have already reached that status (after all, there’s a huge gap between knowing something exists and having the actual demand to use it). Maybe you disagree with this premise and are now about to close the page or read the rest with resentment. That’s your right. But I will reiterate why this model is broken today and what I believe such platforms should look like.

I don’t want to sound too much like a tech insider; I won’t restate “The Innovator’s Dilemma,” but classic cases like Kodak and Blockbuster are quintessential. These companies (and many others) achieved huge success, creating inertia resistant to change. We all know the ending of these stories, but simply shrugging and saying “we can do better” isn’t constructive. So, what exactly caused these outcomes? Do we see signs of this in today’s prediction markets?

Sometimes, the obstacle is technical. Startups often build products in a subjective way that works initially (being able to do so is already overcoming many difficulties!), but quickly becomes a rigid architecture that becomes a future constraint. To continue expanding after initial growth or to adjust product design means threatening some seemingly effective core components. People naturally tend to fix problems through incremental patches, but this quickly turns the product into a patchwork monster. Moreover, this merely delays facing the brutal truth: what is truly needed is a complete rebuild or reimagining of the product.

Early social networks experienced this when reaching performance ceilings. Friendster, a pioneer of 2002 social networks, connected millions of users via “friends of friends.” But when a specific feature (viewing “three degrees of separation” friends) caused the platform to crash under the load of exponential connections, trouble arose.

The team refused to scale down this feature, instead focusing on new ideas and flashy partnerships, even as existing users threatened to defect to MySpace. Friendster reached a local peak of popularity but couldn’t surpass it because its core architecture was flawed, and the team refused to acknowledge, dismantle, and fix it. (By the way, MySpace later fell into its own「local optimum」trap: built around highly customizable user profiles, focusing on music/pop culture groups. The platform was mainly ad-driven, eventually over-reliant on its ad portal model, while Facebook emerged with a cleaner, faster, “real” identity-based network. Facebook attracted some early MySpace users but undoubtedly drew in a larger subsequent social media user base.)

The persistence of such behaviors is unsurprising. We are all human. Achieving some superficial success, especially as a startup with a high failure rate, naturally inflates the ego. Founders and investors begin to believe their hype and double down on their current formula, even as warning signs grow brighter. It’s easy to ignore new information or refuse to face the reality that has already changed from the past. The human brain is fascinating—given enough motivation, we can rationalize many things.

Stagnation of “Research In Motion”

Before the iPhone, Research In Motion (RIM)'s BlackBerry was the king of smartphones, holding over 40% of the US smartphone market. It was built around a specific concept: a better PDA optimized for enterprise users—focused on email, battery life, and that beloved physical keyboard. However…

What might be underestimated today is that BlackBerry was excellent at serving its customers. Because of this, when the world around them changed dramatically, RIM failed to adapt.

It is well known that their leadership initially dismissed the iPhone.

“It’s not secure. The battery drains quickly, and it has a terrible digital keyboard.” — Larry Conlee, RIM COO

They quickly became defensive.

RIM arrogantly believed that this new phone would never appeal to their enterprise customer base, which was not entirely unfounded. But they completely missed the epochal shift where smartphones evolved from “email machines” to “all-purpose devices for everyone.” The company suffered from severe “technical debt” and “platform debt,” common symptoms of early success. Their OS and infrastructure were optimized for secure messaging and battery efficiency. By the time they accepted the reality, it was too late.

Some argue that companies in such situations (the larger the initial success, the harder the evolution—one reason Zuckerberg is considered the “GOAT/Greatest Of All Time”) should operate with a near-split personality: one team leveraging current success, another aiming to disrupt it. Apple might be a paradigm—letting iPhone eat into iPod’s market, then iPad eating into Mac’s. But if it were easy, everyone would do it.

Yahoo

This might be a “Missed Opportunity” on the level of Mount Rushmore. Once, Yahoo was the homepage for millions of internet users. It was the portal (or perhaps the original “all-in-one app”)—news, email, finance, gaming, everything. It regarded search as just one of many functions, to the extent that in the early 2000s, Yahoo didn’t even use its own search tech (outsourcing to third-party engines, even using Google for a time).

Today, it’s well known that Yahoo repeatedly missed opportunities to deepen its search capabilities, most famously the $5 billion acquisition of Google in 2002. It seems obvious in hindsight, but Yahoo failed to understand what Google knew: that search is the foundation of the digital experience. Whoever controls search controls internet traffic and ad revenue. Yahoo relied too heavily on brand strength and display ads, grossly underestimating the shift to「search-centric」navigation and later to social networks with personalized content streams.

Pardon the cliché, but in bubble markets, “water rises with the boat.” Cryptocurrency is no exception (see Opensea and many other examples). It’s hard to tell whether your startup has real traction or is just riding an unsustainable wave of momentum. What complicates matters is that these periods often coincide with a surge in venture capital and speculative consumption, masking underlying fundamental issues. WeWork’s rapid rise and fall exemplifies this: easy capital led to massive expansion, hiding a fundamentally broken business model.

Strip away all branding and highfalutin language, and WeWork’s core business is very simple:

Long-term lease of office space → spend on renovations → short-term premium sublease.

If you’re unfamiliar with this story, you might think, hmm, this sounds like a short-term landlord. That’s exactly what it is—a real estate arbitrage disguised as a software platform.

But WeWork wasn’t necessarily interested in building a lasting enterprise; they optimized for explosive growth and valuation narratives. This worked temporarily because Adam Neumann was charismatic enough to sell the vision. Investors bought in and fueled a growth-at-all-costs mentality (in WeWork’s case, opening as many offices as possible in as many cities as possible, ignoring profitability, believing “we can grow out of losses”). Many outsiders (analysts) saw through it: it was a risk-inverted real estate company with unstable tenants and an inherently structurally unprofitable business.

Most of this is a retrospective analysis of a failed company. In a sense, “Monday morning quarterbacking.” But it reflects three different failure insights: failure due to inability to progress technologically, recognize and respond to competition, or adapt the business model.

I believe we are now witnessing the same pattern in prediction markets.

Prediction Markets’ Promise

The theoretical promise of prediction markets is alluring:

Harnessing collective wisdom = better information = turning speculation into collective insight = infinite markets

But today’s leading platforms have already hit a local peak. They have found a pattern that generates some traction and trading volume, but this design cannot realize the true vision of「all things predictable and sufficiently liquid」.

On the surface, both show signs of success, and no one doubts that. Kalshi reports that the industry’s annualized trading volume will reach about $30 billion this year (more on how much is organic growth later). The industry has seen a new wave of interest in 2024-25, especially as on-chain finance narratives and gamified trading further embed into the cultural zeitgeist. Over-marketing by Polymarket and Kalshi may also be related (sometimes aggressive promotion is effective).

But if we peel back the onion, we find some warning signs that growth and PMF (Product-Market Fit) may not be as they seem. The elephant in the room is liquidity.

For these markets to function, they need deep liquidity—that is, many participants willing to bet on one side, so that prices are meaningful and reveal true price discovery.

Except for a few high-profile markets, Kalshi and Polymarket struggle with this.

Huge trading volumes are concentrated around major events (US elections, Fed decisions), but most markets show wide bid-ask spreads and almost no activity. Often, market makers are reluctant to trade (a recent Kalshi founder admitted their internal market makers are not profitable).

This indicates these platforms have yet to crack the challenge of scaling market breadth and depth. They are stuck at a level: some decent activity in dozens of hot markets, but the「long tail」vision of「markets of all things」remains unfulfilled.

To mask these issues, both rely on incentives and unsustainable behaviors (sounds familiar?), typical of local optima and organic growth limitations (by the way, in this specific market dynamic, I have a feeling most people see these two as the only major competitors).

I don’t think this is necessarily critical at this stage, but if these two teams believe it, then if one is perceived as “leading” in this hypothetical「dual horse race」, it could threaten their survival. An unstable position, based on a flawed assumption in my view.

Polymarket launched a liquidity reward program to narrow spreads (theoretically, placing orders near the current price yields rewards). This helps make order books look tighter and indeed improves trading experience by reducing slippage to some extent. But it’s still a subsidy. Similarly, Kalshi introduced trading volume incentives, essentially paying cash back based on user trading volume. They are spending money to get people to use the product.

Now I can hear some shouting “Uber subsidized for a long time too!!!” Yes, incentives aren’t inherently bad. But that doesn’t mean they are good! (I also find it amusing how people always point to exceptions in rules rather than the pile of dead bodies.) Especially considering the current dynamics of prediction markets, this will soon turn into a hamster wheel that can’t be stopped before it’s too late.

Another fact we need to recognize is that a significant portion of trading volume is fake. I think debating the exact ratio is pointless, but it’s clear that fake trades make markets look more liquid than they really are, as a few participants frequently manipulate the market for profit or to create hype. This means actual demand is weaker than it appears.

“Last Price” Pricing

In a healthy, well-functioning market, you should be able to bet close to the current odds without large price swings. But on these platforms, that’s not the case. Even medium-sized orders can significantly impact odds, indicating insufficient trading volume. These markets tend to reflect only the last trader’s moves, which is at the core of the liquidity problem I mentioned earlier. This situation shows that, although a small core group of users sustains some markets, overall these markets are unreliable and illiquid.

But why is that?

Pure binary markets cannot compete with perpetual contracts. It’s a cumbersome approach that fragments liquidity, and even attempts at workaround solutions are clumsy at best. Many of these markets also feature a strange structure: an “Other” option representing unknown factors, which introduces the problem of splitting emerging competitors into separate markets.

The binary nature also means you cannot offer true leverage in the way users want, which in turn means you cannot generate valuable trading volume like perpetual contracts do. I see endless debates on Twitter about this, but I am still shocked that people fail to recognize: betting $100 on a 1% probability in a prediction market is not the same as opening a 100x leveraged position of $100 on a perpetual exchange.

The hidden secret here is that to solve this fundamental issue, you need to redesign the underlying protocol to allow for generalization and treat dynamic events as first-class citizens. You must create an experience similar to perpetual contracts, which involves addressing the jump risk inherent in binary outcome markets. This is obvious to anyone actively trading perpetuals and prediction markets—and yet, many teams are unaware that these users are precisely the ones you need to attract.

Addressing jump risk requires redesigning the system to ensure asset prices move continuously—that they don’t arbitrarily jump from, say, 45% to 100% (we’ve seen how frequently and blatantly these events are manipulated or insider-traded, but that’s another topic I don’t want to open now. Stop the crime.).

Without solving this core limitation, you will never be able to introduce the leverage needed to make the product attractive to users (those who can bring real value to your platform). Leverage depends on continuous price fluctuations to allow safe liquidation before losses exceed collateral, avoiding sudden jumps (like from 45% to 100%) that wipe out one side of the order book. Without this, you can’t timely add margin or liquidate, and the platform will eventually go bankrupt.

Another key reason these markets don’t work under current structures is the lack of a native multi-outcome hedging mechanism. First, there’s no natural hedge in the current setup because these markets resolve as YES/NO, and the “underlying” is the outcome itself. In contrast, if I go long on BTC perpetuals, I can hedge by shorting BTC elsewhere. This concept doesn’t exist in today’s prediction market structure, so if market makers are forced to bear direct event risk, providing deep liquidity (or leverage) becomes extremely difficult. This again underscores why I believe “prediction markets are a nascent field, in rapid growth” is a naive view.

Prediction markets will eventually settle (i.e., close at resolution), but perpetual futures obviously won’t. They are open-ended. Designing perpetual-like mechanisms can incentivize active trading, making markets more continuous and alleviating some common behaviors that make prediction markets unattractive (many participants just hold until resolution rather than actively trading probabilities). Also, since prediction outcomes are discrete and one-time, while oracle feeds are ongoing (though not without issues), the oracle problem is even more prominent in prediction markets.

Behind these design issues are capital efficiency problems, which are now quite evident. I personally believe that earning stablecoins with invested capital doesn’t bring substantial change. Especially since exchanges will offer such yields anyway. So what’s the trade-off here? If every trade is fully prepaid, it’s good for eliminating counterparty risk! And you can attract some users.

But it’s disastrous for the broader user base you need; this model is highly inefficient from a capital perspective and greatly increases participation costs. When these markets require different types of users to operate at scale, that’s especially bad, because these choices worsen the experience for each user group. Market makers need large capital pools to provide liquidity, while retail traders face huge opportunity costs.

There’s undoubtedly more to unpack, especially around how to attempt solving some of these fundamental challenges. More sophisticated and dynamic margin systems will be necessary, especially considering factors like “time to event” (when the event resolution is near and odds are close to 50/50, risk is highest). Introducing concepts like leverage decay near resolution and tiered liquidation levels early on will also help.

Borrowing from traditional finance broker models, implementing instant collateralization is another step in the right direction. It frees up capital, allows for more efficient use, and enables cross-market orders with updates after execution. Starting with scalar markets and then expanding to binary markets seems the most logical sequence.

The key point is that there’s a vast space of design options yet to be explored, partly because many believe today’s models are already final. I simply don’t see enough people willing to first confront these limitations head-on. Perhaps it’s not surprising that those who realize this are often the types of users these platforms should aim to attract (aka perpetual contract traders).

But what I see is that most criticisms of prediction markets are dismissed by supporters, who are told to look at the trading volume and growth figures of these two platforms (absolutely real and organic numbers, hmm). I hope prediction markets develop, that they become widely accepted, and I personally think that a market of all things is a good idea. Most of my frustration stems from a common belief that today’s version is the best, but I clearly disagree.